1 - (SYM 15) Predicting ADHD Symptom Change Using Wearable-derived Activity and Sleep Features: A Secondary Analysis of Randomized Trial Data

Attention-deficit/hyperactivity disorder (ADHD) in adults affects their academic, occupational, and daily functioning, and fluctuations in symptoms contribute to ongoing difficulties in everyday life. Predicting symptom changes using objective indicators may support early identification of deterioration and inform tailored support. This study aims to construct predictive models of symptom change using wearable data collected during a 12-week randomized controlled trial in which adults with ADHD were allocated to videoconference-based cognitive behavioral therapy or a treatment as usual condition. While exploratory comparisons between the intervention groups were conducted, the primary focus of this analysis was to examine the predictive utility of wearable activity and sleep features for symptom changes.

During the 12-week trial period, participants completed 12 treatment or control sessions and wore wearable devices that collected heart rate and sleep-related data. Valid wearable and symptom data were available for 28 participants, yielding 12 repeated measurements per participant. Circadian rhythm features were derived from heart rate time series, including average level (mesor), amplitude, and phase indicators reflecting peak activity timing. Features were calculated over the three days preceding each session and expressed as differences from the initial session. These features were used to build a machine learning model (LightGBM) to predict changes in ADHD symptom severity, with feature contributions evaluated using Shapley additive explanations. As a secondary analysis, sleep regularity was assessed using the Sleep Regularity Index (SRI) and examined in relation to the ADHD symptom measures across the same 12 measurement points.

Baseline ADHD symptom measures contributed substantially to the model output, while heart rate–derived circadian features independently complemented symptom change prediction. In the secondary analysis, mean SRI preceding each assessment was significantly associated with the ADHD symptom scores among the 28 participants with complete data.

These findings indicate that low-burden, continuously monitored wearable indicators may serve as promising sources of information for predicting changes in symptoms in adults with ADHD. Further refinement of predictive algorithms and integration of continuous monitoring may enable the characterization of symptom trajectories, early detection of worsening symptoms, and support decisions regarding the timing of interventions.